Ride-hailing companies like OLA face a major challenge with high driver churn. Recruiting new drivers is costly, and retaining existing ones is far more sustainable. The goal is to identify whether a driver is likely to leave the company based on attributes such as:
- Demographics (city, age, gender)
- Tenure details (joining date, last working date)
- Performance history (monthly business value, quarterly rating, grade, income)
Build and evaluate ensemble learning models to predict driver attrition (whether a driver will leave OLA) using historical driver-level data.
The project aims to:
- Perform data cleaning and preprocessing
- Explore patterns influencing driver churn
- Build machine-learning models using bagging, boosting, stacking, etc.
- Compare model performance to identify the best approach
- Generate actionable insights to help reduce driver churn
| Column | Description |
|---|---|
| MMMM-YY | Reporting Date (Monthly) |
| Driver_ID | Unique id for drivers |
| Age | Age of the driver |
| Gender | Gender of the driver – Male : 0, Female: 1 |
| City | City Code of the driver |
| Education_Level | Education level – 0 for 10+ ,1 for 12+ ,2 for graduate |
| Income | Monthly average Income of the driver |
| Date Of Joining | Joining date for the driver |
| LastWorkingDate | Last date of working for the driver |
| Joining Designation | Designation of the driver at the time of joining |
| Grade | Grade of the driver at the time of reporting |
| Total Business Value | The total business value acquired by the driver in a month (negative business indicates cancellation/refund or car EMI adjustments) |
| Quarterly Rating | Quarterly rating of the driver: 1,2,3,4,5 (higher is better) |
- Handling missing values
- Converting string date fields to datetime objects
- Dropping irrelevant columns (e.g., Unnamed: 0)
- Encoding categorical features
- Feature engineering (e.g., tenure calculation)
- Distribution analysis of churn vs non-churn
- Relationship between income, ratings, tenure, and churn
- Outlier detection and treatment
- Logistic Regression (baseline)
- Bagging: Random Forest, Extra Trees
- Boosting: Gradient Boosting, XGBoost, AdaBoost, CatBoost
- Stacking / Voting Ensemble
- Hyperparameter tuning
- Accuracy
- Recall / Sensitivity (important for churn detection)
- Precision
- F1-Score
- ROC/AUC
- Dataset contains 19,104 records and 14 features.
- Gender, education level, and city require categorical encoding.
- Business performance variables (e.g., Total Business Value) show strong correlation with attrition.
- Drivers with low quarterly ratings, lower income, and shorter tenure churn at higher rates.
- Drivers with negative business value are significantly more likely to leave.
- Date fields contain inconsistent formats and required corrections.
- Tenure is one of the strongest predictors of whether a driver will leave.
- Quarterly rating positively correlates with retention — low-rated drivers leave more often.
- Income fluctuations contribute heavily to churn.
- Ensemble models outperform classical models in handling non-linear churn patterns.
- Boosting methods (particularly XGBoost/CatBoost) deliver the highest predictive power.
- Implement a driver retention dashboard using the model’s output to identify high-risk drivers early.
- Introduce incentive programs for low-income and low-rated drivers.
- Monitor first-90-days tenure closely — targeted support here can reduce early attrition.
- Use business-value trends to proactively engage drivers whose performance is declining.
- Perform periodic model retraining as new driver behavior data becomes available.
This project successfully demonstrates how ensemble learning methods can be leveraged to predict driver attrition with high accuracy. Boosting models such as XGBoost/CatBoost performed best, offering actionable insights to help OLA:
- Reduce the cost of hiring
- Increase driver stability
- Improve customer service indirectly through reliable supply
- Make data-driven policy decisions